Neural network based control for UUV with deployment on the High-level Synthesis

Abstract

With the improvement of marine technology, unmanned underwater vehicle (UUV) plays an important role in marine survey operations, environment monitoring and so on. In practical applications, the neural network (NN) based controller is able to compensate for the uncertain and nonlinear disturbance caused by UUV dynamic model. However, due to the performance limitation of embedded hardware, it is still a challenge to deploy the NN based controller. In this paper, Field Programmable Gate Array (FPGA) with parallel computing ability is used to realize the deployment of NN based controller. The High-level Synthesis (HLS) technology is utilized to accelerate the computation of the NN. The hardware in loop simulations show that the longest computing time on FPGA platform is about 6.51us, which is far less than the maximum 1400us computing time on raspberry pi platform. The average relative error of the controller output deployed on FPGA and Raspberry pi is less than 0.5%. The design method not only improves the calculation speed, but also guarantee high real-time response in the UUV control task.